1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
|
; HP-PA-1.1 __mpn_mul_1 -- Multiply a limb vector with a limb and store
; the result in a second limb vector.
; Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc.
; This file is part of the GNU MP Library.
; The GNU MP Library is free software; you can redistribute it and/or modify
; it under the terms of the GNU Library General Public License as published by
; the Free Software Foundation; either version 2 of the License, or (at your
; option) any later version.
; The GNU MP Library is distributed in the hope that it will be useful, but
; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
; License for more details.
; You should have received a copy of the GNU Library General Public License
; along with the GNU MP Library; see the file COPYING.LIB. If not, write to
; the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
; INPUT PARAMETERS
; res_ptr r26
; s1_ptr r25
; size r24
; s2_limb r23
; This runs at 9 cycles/limb on a PA7000. With the used instructions, it can
; not become faster due to data cache contention after a store. On the
; PA7100 it runs at 7 cycles/limb, and that can not be improved either, since
; only the xmpyu does not need the integer pipeline, so the only dual-issue
; we will get are addc+xmpyu. Unrolling would not help either CPU.
; We could use fldds to read two limbs at a time from the S1 array, and that
; could bring down the times to 8.5 and 6.5 cycles/limb for the PA7000 and
; PA7100, respectively. We don't do that since it does not seem worth the
; (alignment) troubles...
; At least the PA7100 is rumored to be able to deal with cache-misses
; without stalling instruction issue. If this is true, and the cache is
; actually also lockup-free, we should use a deeper software pipeline, and
; load from S1 very early! (The loads and stores to -12(sp) will surely be
; in the cache.)
.code
.export __mpn_mul_1
__mpn_mul_1
.proc
.callinfo frame=64,no_calls
.entry
ldo 64(%r30),%r30
fldws,ma 4(%r25),%fr5
stw %r23,-16(%r30) ; move s2_limb ...
addib,= -1,%r24,L$just_one_limb
fldws -16(%r30),%fr4 ; ... into fr4
add %r0,%r0,%r0 ; clear carry
xmpyu %fr4,%fr5,%fr6
fldws,ma 4(%r25),%fr7
fstds %fr6,-16(%r30)
xmpyu %fr4,%fr7,%fr8
ldw -12(%r30),%r19 ; least significant limb in product
ldw -16(%r30),%r28
fstds %fr8,-16(%r30)
addib,= -1,%r24,L$end
ldw -12(%r30),%r1
; Main loop
L$loop fldws,ma 4(%r25),%fr5
stws,ma %r19,4(%r26)
addc %r28,%r1,%r19
xmpyu %fr4,%fr5,%fr6
ldw -16(%r30),%r28
fstds %fr6,-16(%r30)
addib,<> -1,%r24,L$loop
ldw -12(%r30),%r1
L$end stws,ma %r19,4(%r26)
addc %r28,%r1,%r19
ldw -16(%r30),%r28
stws,ma %r19,4(%r26)
addc %r0,%r28,%r28
bv 0(%r2)
ldo -64(%r30),%r30
L$just_one_limb
xmpyu %fr4,%fr5,%fr6
fstds %fr6,-16(%r30)
ldw -16(%r30),%r28
ldo -64(%r30),%r30
bv 0(%r2)
fstws %fr6R,0(%r26)
.exit
.procend
|
GitWeb